In the last decade, the growth of the micro-industry in urban areas has produced an increase in the frequency of xenobiotic polluting discharges in drainage systems. Wastewater treatment plants are usually characterized by low removal efficiencies in respect of such pollutants, which may have an acute or cumulative impact on environmental and public health. To facilitate the early isolation of illicit intrusions, this study aims to develop an approach for positioning water quality sensors based on the Bayesian decision network (BDN). The analysis is focused on soluble conservative pollutants, such as metals. The proposed methodology incorporates several sources of information, including network topology, flows and non-formal ‘grey’ information about the possible locations of contamination sources. The methodology is tested using two sewer systems with increasing complexity: a literature scheme from the Storm Water Management Model (SWMM) manual and a real combined sewer in Italy. In both cases, the approach identifies the optimal sensor location gaining advantage from additional information, which reduces the computational effort needed to obtain the solution. In the real case, the application of the method yielded a better solution with regards to the real position of the implemented sensor network.